The planning of metro-led underground spaces often neglects public perception, while traditional methods struggle to translate unstructured social media data into actionable insights. This study introduces a novel framework that integrates large language models (LLMs) and machine learning to quantify public experience systematically. Using 25,349 Google Maps reviews for 91 Hong Kong metro stations, an LLM engine performs fine-grained perception extraction across 14 indicators. We propose a “perception value” (PV) metric that synthesizes perception frequency and preference, significantly improving modeling performance. Machine learning analysis identifies “transfer” as the most critical contribution to public satisfaction. Notably, “barrier-free design” emerges as a high-contribution but low-preference factor, signaling an urgent need for improvement. Spatially, our analysis reveals a distinct “center-periphery” pattern in perception quality, enabling targeted renewal strategies. Our framework provides a scalable, interpretable paradigm for transforming public perception into planning intelligence, advancing human-centered governance for urban underground spaces.
The master planning of urban underground space (UUS) is essential for guiding orderly development, spatial coordination, and sustainable resource allocation. However, existing UUS planning remains largely static and blueprint-oriented, with insufficient attention to the spatial equilibrium among development demand, supply capacity, and existing utilization performance. Taking the central urban area of Jinan, China, as a case study, this research develops a data-driven spatial equilibrium framework based on multi-source urban data. A multi-dimensional spatial autocorrelation analysis matrix (MSAAM) is constructed by integrating bivariate spatial autocorrelation analysis (BSAA) results to identify local–neighbor relationships and differentiated UUS development strategies. The identified combinations are classified into five types: Internal Development, External Development, Balanced Development, Limited Development, and Self-contained Development. Among the 869 classified spatial units, Limited Development accounts for the largest proportion (57.5 %), followed by Balanced Development (23.2 %), Internal Development (7.8 %), Self-contained Development (7.5 %), and External Development (3.9 %). Balanced Development areas are concentrated in mature built-up areas, whereas Internal and External Development areas occur mainly in urban expansion and transitional zones. Limited and Self-contained Development areas are primarily located in urban–rural fringe areas, ecological corridors, and low-density peripheral zones. These findings show that UUS development is shaped by spatial differences in the alignment of demand, supply, and benefit. The framework supports differentiated master planning, strategic zoning, development prioritization, and more efficient allocation of underground space resources.
Artificial intelligence (AI) is beginning to enter preschool and early primary classrooms, offering the potential for personalised learning, inclusive support, and reduced teacher workload. However, this nascent uptake raises acute questions regarding privacy, bias, and children’s developmental well-being. This conceptual article synthesises five foundational educational theories to evaluate AI’s potential affordances and risks for learners aged 3–8. Drawing on a purposive sample of illustrative studies, design reports, and policy documents (2015–2025), five emerging application areas are identified: personalised curricula, inclusive supports, early screening, augmented play, and teacher empowerment. These applications are mapped onto a proposed Holistic Ethical Integration Model, which assigns interdependent responsibilities to designers, educators, parents, and regulators. The model yields design principles and policy recommendations aimed at guiding responsible adoption while safeguarding young children’s rights. Instead of an exhaustive literature review, this theory-driven roadmap lays the groundwork for future empirical testing and evidence-based policymaking.
In the context of globalization and rapid technological advancement, traditional supply chains face mounting pressure to address environmental challenges like high energy use, waste, and carbon emissions. Growing sustainability awareness compels firms to innovate. Artificial intelligence (AI) has emerged as a key enabler of green supply chain transformation, offering advanced data processing, pattern recognition, and decision optimization to enhance both efficiency and sustainability. This study examines how AI-driven green innovation affects sustainable development across economic, environmental, and social dimensions. Using a mixed-methods approach—comprising a literature review, enterprise surveys, and case studies—the research quantifies AI’s benefits, including cost savings, lower emissions, and reduced waste, while identifying barriers such as high implementation costs, poor partner coordination, and inadequate infrastructure. Findings indicate that AI significantly advances sustainability by optimizing logistics, improving demand forecasting, cutting energy use, and enabling integration with technologies like blockchain and IoT to boost transparency and resilience. The study contributes theoretically by clarifying how AI fosters green innovation and offers practical guidance for businesses and policymakers to support AI adoption in pursuit of sustainable development goals. Despite limitations like potential sampling bias, it highlights AI’s transformative role in building sustainable supply chains.
Soft dual-modal tactile sensors capable of simultaneously sensing force and temperature are essential for enhancing human-like perception and interaction in robots, particularly in the functional sense of concurrent mechanical and thermal perception. However, achieving self-decoupled and high-fidelity dual-modal sensing remains a significant challenge due to intrinsic signal crosstalk, structural complexity, and limited flexibility in existing designs. Here, we present a soft robotic tactile (RoboTac) skin that intrinsically decouples force and temperature using an ionic conductive film within a minimalist architecture, featuring an ultralight weight and an ultralow cost. Ionic conductivity enables independent readouts without algorithmic compensation by allowing thickness compression to modulate capacitance (force) and lateral ionic transport under thermal stimuli to modulate resistance (temperature). Moreover, the RoboTac skin demonstrates its practical utility for robots in object perception, specialized tasks, and human-robot interaction. This work establishes a general principle for intrinsically self-decoupling modalities in tactile sensors, advancing multimodal sensing, intelligent perception, and embodied robotics.
